Search

CN-119835219-B - Traffic forwarding method and device and related equipment

CN119835219BCN 119835219 BCN119835219 BCN 119835219BCN-119835219-B

Abstract

The present application relates to the technical field of intelligent computing centers, and in particular, to a traffic forwarding method, apparatus and related devices. The method comprises the steps of determining a target path for transmitting traffic of a task to be trained based on a source destination address of the task to be trained, determining a target outlet interface for forwarding the traffic of the task to be trained on a target Leaf accessed by a source end server, sending information for requesting a target source port value to the target Leaf, calculating a corresponding target source port value by the target Leaf based on the source destination address and the target outlet interface of the task to be trained, and feeding the target source port value back to a controller, receiving the target source port value fed back by the target Leaf, and indicating the source end server to set a source port of the traffic of the task to be trained as the target source port value when the task to be trained is executed, wherein the target Leaf hashes the traffic of the task to be trained to the target outlet interface for forwarding based on the source destination address and the target source port value of the traffic of the task to be trained.

Inventors

  • YU YANMIN

Assignees

  • 新华三技术有限公司

Dates

Publication Date
20260505
Application Date
20241219

Claims (8)

  1. 1. A method for forwarding traffic, the method being applied to a controller in a Spine-to-Leaf Spine network, the method comprising: Determining a target path for transmitting the traffic of the task to be trained based on a source and destination address of the task to be trained, and determining a target outlet interface for forwarding the traffic of the task to be trained on a target Leaf accessed by a source end server based on the target path; Sending information for requesting a target source port value to the target Leaf, wherein after receiving the information, the target Leaf calculates a corresponding target source port value based on a source destination address of the task to be trained and the target output interface, and feeds back the target source port value to the controller; Receiving a target source port value fed back by the target Leaf, and indicating the source end server to set a message source port of the to-be-trained task flow as the target source port value when the to-be-trained task is executed, wherein the target Leaf hashes the to-be-trained task flow to the target output interface for forwarding based on a source destination address of the to-be-trained task flow and the target source port value; The step of determining a target path for transmitting traffic of the task to be trained based on a source destination address of the task to be trained comprises: Acquiring the topology information of the whole network, wherein the topology information of the whole network comprises the relationship of links and the information of access interfaces included in the links; defining initial overhead values of links included in the whole network topology information, wherein the size of the initial overhead value of one link is inversely proportional to the maximum bandwidth value of the link; according to source and destination end information of a task to be trained, host ARP information, mapping relations between all the ingress ports included by all the port groups constructed on all the Spine equipment and all the equivalent routes included by all the equivalent route groups and overhead values of all the links included by the whole network topology information, determining a target path for transmitting the traffic of the task to be trained, wherein the overhead values of all the target links included by the target path are smaller than or equal to preset values.
  2. 2. A traffic forwarding method, applied to an origin server, the method comprising: Receiving an indication that a message source port of a task to be trained is set as a target source port value, wherein the indication is sent by a controller, the controller determines a target path for transmitting the flow of the task to be trained based on a source destination address of the task to be trained, determines a target outlet interface for forwarding the flow of the task to be trained on a target Leaf accessed by a source end server based on the target path, sends information for requesting the target source port value to the target Leaf, and after receiving the information, calculates a corresponding target source port value based on a source destination address of the task to be trained and the target outlet interface and feeds back the target source port value to the controller; the method comprises obtaining full network topology information, host ARP information and mapping relations between each ingress port included in each port group constructed on the Spine equipment and each equivalent route included in each equivalent route group, wherein the full network topology information comprises each link relation and ingress and egress interface information included in each link, defining initial overhead value of each link included in the full network topology information, wherein the initial overhead value of one link is inversely proportional to the maximum bandwidth value of the link, determining the target path for transmitting the flow of the task according to the source and destination end information of the task to be trained, host ARP information, mapping relations between each ingress port included in each port group constructed on each Spine equipment and each equivalent route included in each equivalent route group and overhead value of each link included in the full network topology information, the sum of the overhead values of all target links included in the target path is smaller than or equal to a preset value; And sending the task flow to be trained, of which the message source port is the target source port value, to the target Leaf, so that the target Leaf hashes the task flow to be trained to the target output interface for forwarding based on the source destination address of the task flow to be trained and the target source port value.
  3. 3. The method of claim 2, wherein the source server has installed thereon a proxy program that receives an indication sent by a controller to set a message source port of a task to be trained to a target source port value, sets the message source port of the task traffic to be trained to the target source port value when it is determined to start the task to be trained, and sends the task traffic to be trained, with the message source port being the target source port value, to the target Leaf.
  4. 4. A traffic forwarding device for use in a controller in a Spine-to-Leaf Spine network, the device comprising: the system comprises a determining unit, a source-destination address determining unit and a source-destination address determining unit, wherein the determining unit is used for determining a target path for transmitting the flow of a task to be trained based on the source-destination address of the task to be trained, and determining a target outlet interface for forwarding the flow of the task to be trained on a target Leaf accessed by a source-end server based on the target path; The sending unit is used for sending information for requesting a target source port value to the target Leaf, wherein after the target Leaf receives the information, the corresponding target source port value is calculated based on a source destination address of the task to be trained and the target output interface, and the target source port value is fed back to the controller; The receiving unit is used for receiving a target source port value fed back by the target Leaf and indicating the source end server to set a message source port of the flow of the task to be trained as the target source port value when the task to be trained is executed, wherein the target Leaf hashes the flow of the task to be trained to the target output interface for forwarding based on a source destination address of the flow of the task to be trained and the target source port value; When determining a target path for transmitting traffic of a task to be trained based on a source and destination address of the task to be trained, the determining unit is specifically configured to: Acquiring the topology information of the whole network, wherein the topology information of the whole network comprises the relationship of links and the information of access interfaces included in the links; defining initial overhead values of links included in the whole network topology information, wherein the size of the initial overhead value of one link is inversely proportional to the maximum bandwidth value of the link; according to source and destination end information of a task to be trained, host ARP information, mapping relations between all the ingress ports included by all the port groups constructed on all the Spine equipment and all the equivalent routes included by all the equivalent route groups and overhead values of all the links included by the whole network topology information, determining a target path for transmitting the traffic of the task to be trained, wherein the overhead values of all the target links included by the target path are smaller than or equal to preset values.
  5. 5. A traffic forwarding device for use with an origin server, the device comprising: The receiving unit is used for receiving an instruction sent by the controller to set a message source port of a task to be trained as a target source port value, wherein the controller determines a target path for transmitting the traffic of the task to be trained based on a source destination address of the task to be trained, determines a target outlet interface which is accessed by the source end server and is used for forwarding the traffic of the task to be trained based on the target path, and then sends information for requesting the target source port value to the target Leaf, and after receiving the information, the target Leaf calculates a corresponding target source port value based on a source destination address of the task to be trained and the target outlet interface and feeds back the target source port value to the controller; the method comprises obtaining full network topology information, host ARP information and mapping relations between each ingress port included in each port group constructed on the Spine equipment and each equivalent route included in each equivalent route group, wherein the full network topology information comprises each link relation and ingress and egress interface information included in each link, defining initial overhead value of each link included in the full network topology information, wherein the initial overhead value of one link is inversely proportional to the maximum bandwidth value of the link, determining the target path for transmitting the flow of the task according to the source and destination end information of the task to be trained, host ARP information, mapping relations between each ingress port included in each port group constructed on each Spine equipment and each equivalent route included in each equivalent route group and overhead value of each link included in the full network topology information, the sum of the overhead values of all target links included in the target path is smaller than or equal to a preset value; and the sending unit is used for sending the task flow to be trained, of which the message source port is the target source port value, to the target Leaf, so that the target Leaf hashes the task flow to be trained to the target output interface for forwarding based on the source destination address of the task flow to be trained and the target source port value.
  6. 6. The apparatus of claim 5 wherein an agent is installed on the source server, the agent receiving an indication sent by a controller to set a message source port of a task to be trained to a target source port value, and upon determining to launch the task to be trained, setting the message source port of the task traffic to be trained to the target source port value, and sending the task traffic to be trained, with the message source port being the target source port value, to the target Leaf.
  7. 7. A traffic forwarding device, the traffic forwarding device comprising: A memory for storing program instructions; A processor for invoking program instructions stored in said memory and for performing the steps of the method according to any of claims 1-3 in accordance with the obtained program instructions.
  8. 8. A computer readable storage medium storing computer executable instructions for causing a computer to perform the steps of the method according to any one of claims 1-3.

Description

Traffic forwarding method and device and related equipment Technical Field The present application relates to the technical field of intelligent computing centers, and in particular, to a traffic forwarding method, apparatus and related devices. Background In an intelligent data center network, the uplink bandwidth and the downlink bandwidth are 1:1, more equivalent-Cost Multi-Path routing (ECMP) links exist in the networking, under the condition of stream-by-stream forwarding, a switch performs hash routing forwarding according to five-tuple and other calculation factors of traffic, load balancing in the network is realized, and traffic of a plurality of inlets hash ECMP to the same outlet. Other outlets have no traffic, resulting in load sharing imbalance and network congestion. That is, in the intelligent data center networking, the switch devices are hashed or loaded unevenly, which results in network traffic congestion that cannot be resolved. Disclosure of Invention The application provides a traffic forwarding method, a traffic forwarding device and related equipment. In a first aspect, the present application provides a traffic forwarding method, applied to a controller in a Spine-Leaf Spine network, the method comprising: Determining a target path for transmitting the traffic of the task to be trained based on a source and destination address of the task to be trained, and determining a target outlet interface for forwarding the traffic of the task to be trained on a target Leaf accessed by a source end server based on the target path; Sending information for requesting a target source port value to the target Leaf, wherein after receiving the information, the target Leaf calculates a corresponding target source port value based on a source destination address of the task to be trained and the target output interface, and feeds back the target source port value to the controller; Receiving a target source port value fed back by the target Leaf, and indicating the source end server to set a message source port of the to-be-trained task flow as the target source port value when the to-be-trained task is executed, wherein the target Leaf hashes the to-be-trained task flow to the target output interface for forwarding based on a source destination address of the to-be-trained task flow and the target source port value. Optionally, the step of determining a target path for transmitting traffic of the task to be trained based on a source destination address of the task to be trained includes: Acquiring the topology information of the whole network, wherein the topology information of the whole network comprises the relationship of links and the information of access interfaces included in the links; defining initial overhead values of links included in the whole network topology information, wherein the size of the initial overhead value of one link is inversely proportional to the maximum bandwidth value of the link; according to source and destination end information of a task to be trained, host ARP information, mapping relations between all the ingress ports included by all the port groups constructed on all the Spine equipment and all the equivalent routes included by all the equivalent route groups and overhead values of all the links included by the whole network topology information, determining a target path for transmitting the traffic of the task to be trained, wherein the overhead values of all the target links included by the target path are smaller than or equal to preset values. In a second aspect, the present application provides a traffic forwarding method, applied to a source end server, where the method includes: Receiving an indication that a message source port of a task to be trained is set as a target source port value, wherein the indication is sent by a controller, the controller determines a target path for transmitting the flow of the task to be trained based on a source destination address of the task to be trained, determines a target outlet interface for forwarding the flow of the task to be trained on a target Leaf accessed by a source end server based on the target path, sends information for requesting the target source port value to the target Leaf, and after receiving the information, calculates a corresponding target source port value based on a source destination address of the task to be trained and the target outlet interface and feeds back the target source port value to the controller; And sending the task flow to be trained, of which the message source port is the target source port value, to the target Leaf, so that the target Leaf hashes the task flow to be trained to the target output interface for forwarding based on the source destination address of the task flow to be trained and the target source port value. Optionally, an agent is installed on the source server, and the agent receives an indication that a message source port of a task to be trained is set